There is one positive thing I saw in this call. On page 12 of the deck there is a matrix comparing the effectiveness of Upstart risk grade and FICO.
For FICO the default rate ranges between 3.4 to 7.7
For Upstart it ranges between 0.7 and 9.0
If the chart is correct with a large sample size, then this is a great technical achievement - in machine learning, we call this type of classification model “discriminative” - the goal is to find sharp boundaries and push the good ones toward one end and bad ones toward the other as far as possible. A typical example is that a weather forecast model that puts out “50% chance of rain” every day is much less useful and actionable than a discriminative one that puts out either “5% chance of rain” and “95% change of rain” for daily forecasts. I think this is the right metric to chase and shows the power of product.
A more discriminative model means there are more levers for controlling risk and pricing the loans. This is very good.
The only problem with this chart is that I can’t see the sample size for each of the vintage (they could be cherry-picking the FICO segments) and it also doesn’t show year to year trend on whether the model performance has degraded or not recently.
On the other hand, besides the clear revenue problem, another bad I saw in the call was that they said the rising interest rates will push some applications above the 36% interest rate threshold that are set by state laws as the maximum. This was a blind spot for me as I didn’t think of this issue in my investment thesis. I interpret it as that they will lose the “riskiest” segment of revenue for now until their model improves. Given that rates will probably stay high for some time, this can really take a while. The lost revenue, in my opinion, has to come from other products that are currently in their infancies and will not be material before 2023.
The stalling revenue is so bad that it puts Upstart square in the “story stock” zone for me. Given the high growth of other companies, even “holding for one year” with little to no growth in Upstart’s revenue means I’ll be losing out on an almost guaranteed 60-70% revenue increase from other high quality SaaS companies. Upstart is certainly a complex story compared to SaaS and I have learned a pricy lesson.